Automated Coronal Hole Detection using He I 1083 nm Spectroheliograms and Photospheric Magnetograms

0.58 MB·English

by  C. J. Henney

**FULL TITLE**
ASP Conference Series, Vol. **VOLUME**, **YEAR OF PUBLICATION**
**NAMES OF EDITORS**
Automated Coronal Hole Detection using He I 1083 nm
Spectroheliograms and Photospheric Magnetograms
C. J. Henney and J. W. Harvey
National Solar Observatory, Tucson, Arizona, USA
7
0 Abstract. AmethodforautomatedcoronalholedetectionusingHeI1083nm
0
spectroheliogramsandphotosphericmagnetogramsispresentedhere. Theunique
2
line formation of the helium line allows for the detection of regions associated

n with solar coronal holes with minimal line-of-sight obscuration across the ob-
a served solar disk. The automated detection algorithm utilizes morphological
J
image analysis, thresholding and smoothing to estimate the location, bound-

5 aries,polarityandflux ofcandidatecoronalholeregions. Thealgorithmutilizes
thresholdsbasedonmeanvaluesdeterminedfromover10yearsoftheKittPeak
1
Vacuum Telescope daily hand-drawn coronal hole images. A comparison be-
v
tween the automatically created and hand-drawn images for a 11-year period
2
beginning in 1992 is outlined. In addition, the creation of synoptic maps using
2
the daily automated coronal hole images is also discussed.
1
1
0
7
0
/
h 1. Introduction
p
-
The association of high-speed solar wind with solar coronal holes is of great
o
r importance for forecasting space weather events (e.g. Zirker 1977; Arge et al.
t
s 2004). Coronal holes are low density regions in the solar corona that have less
a
extreme ultraviolet and X-ray emission than nominal quiet and active regions.
:
v Themagnetic fields within these regions are mostly unipolarand extend beyond
Xi thecorona into theinterplanetary medium(e.g. Bohlin 1977). For ground-based
observations, a useful proxy for estimating the location of coronal holes is the
r
a He I 1083 nm line. The unique He I 1083 nm line formation in the chromo-
sphere results in decreased absorption (e.g. Andretta & Jones 1997). The Kitt
Peak Vacuum Telescope (KPVT) He I 1083 nm equivalent width images (e.g.
Harvey & Sheeley 1977) are defined such that these regions appear bright (see
Figure 1). KPVT measurements of the sun at He I 1083 nm span from 1974
throughSeptember2003(e.g.Harvey & Livingston1994). AsampleKPVTHeI
1083 nm spectroheliogram, along with a comparison Extreme Ultraviolet Imag-
ing Telescope (EIT) 19.5 nm Fe XII emission line image are shown in Figure 1
(note that the coronal hole regions appear dark in the EIT image).
Using KPVT He I 1083 nm spectroheliograms and photospheric magne-
tograms, daily computer-assisted “hand drawn” (HD) coronal hole images by
Karen Harvey and Frank Recely span from April 1992 through September 2003.
In addition, they constructed coronal hole Carrington synoptic maps for the
period between September 1987 and March 2002 (see Harvey & Recely 2002).
Prior to these periods HD maps were drawn on photographic prints and incor-
porated in synoptic maps published in Solar Geophysical Data. The creation of
1
2 Henney & Harvey
NSO/KP CORONAL HOLE MAP: HE I 1083 nm
Preliminary N Bo=+4.2
E W
07/14/03 S 1705 UT
Figure 1. KPVT He I 1083 nm spectroheliogram (top, left), EIT 19.5 nm
Fe XII emission line image (top, right), the KPVT associated “hand-drawn”
coronal hole image (bottom, left), and the automated detection coronal hole
image (bottom, right) for July 14, 2003 at approximately 17 UT.
theHDcoronalholeimagesandsynopticmapsisnotablylaborintensive, requir-
ing iterative visual inspection of high resolution helium spectroheliograms and
photospheric magnetograms. The production of HD coronal hole maps stopped
with the conclusion of operation of the KPVT in September 2003. Unpublished
HD daily and synoptic maps exist for periods back to 1974.
The motivation for developing an automated detection (AD) coronal hole
algorithm is tocreatedaily images andsynopticmapssimilar tothehand-drawn
maps for the period between 1974 and 1992 using available KPVT data. In ad-
dition, the AD algorithm will be used to create coronal hole images using the
dailySOLISVector SpectroMagnetograph (VSM)heliumspectroheliograms and
photospheric magnetograms. The SOLIS-VSM began synoptic observations in
August2003. InitiallyduringthedevelopmentoftheADcoronalholealgorithm,
only the helium equivalent width images were utilized, at various spatial scales
and thresholds, with limited success (Henney & Harvey 2001). The critical ad-
dition for the detection recipe presented here is the inclusion of magnetograms
Automated Coronal Hole Detection 3
in the analysis to determine the percent unipolarity of the candidate region
(e.g. Harvey & Recely 2002) to avoid areas of polarity inversions and centers of
supergranular cells that otherwise resemble coronal holes.
Other methods for detecting coronal holes include utilizing spectral line
propertiesofHeI1083nm(Malanushenko & Jones 2005)andmulti-wavelength
analysis (de Toma & Arge 2005). In future work, we plan to compare these
methods in detail with the AD coronal hole algorithm presented here. The
following sections outline the parameterizing of coronal holes (Section 2), the
coronal hole detection recipe, along with a comparison with the HD coronal
hole images and the creation of synoptic maps (Section 3).
2. Coronal Hole Parameters
The preliminary step in the development of the automated coronal hole de-
tection algorithm outlined in the following section is to parameterize unique
properties of coronal holes. This allows for general threshold limits to be set
for the input data. The thresholds used to select coronal hole candidate regions
are estimated using successive pairs of KPVT magnetogram and helium spec-
troheliogram data originally used to create the HD coronal hole maps. Using
the coronal hole boundaries from the hand drawn images, the area of each coro-
nal hole is projected into the corresponding average magnetogram and helium
spectroheliogram image in heliographic coordinates.
Theaverage images are created with two consecutive observations weighted
inversely by one plus the time difference between the observations in fractional
days. Thetwo images are averaged in heliographic coordinates sine-latitude and
longitude, where the older image is rotated into the time reference frame of the
most recent image. The KPVT data coverage for the period April 1992 through
September 2003 includes 2,781 image pairs. For this period, the boundaries of
11,241 HD coronal holes were used to estimate the mean percent unipolarity for
these regions relative to latitude and central meridian distance.
InFigure2, thenumberofcoronalholes (top), thenormalizedHeI1083 nm
equivalent width (middle) and the mean percent unipolarity (bottom) are ex-
hibited for the period between 1992 and 2003. The helium equivalent width is
normalized by the median of all positive values, where the zero point is essen-
tially the mean of the disk values excluding active regions and filaments. Note
that for the KPVT helium equivalent width images the limb darkening is re-
moved. For this period a greater number of coronal holes were detected in the
southern hemisphere. The difference between the hemispheres could be a result
of a delayed or slower migration of the holes to the southern polar region or
possibly longer lived regions, since no attempt was made to avoid recounting
long-lived regions. Thegreater numberof detections on the eastern limb is most
likely a result of false detections. For the normalized helium equivalent width
and the mean unipolarity percentage shown in Figure 2, the dashed lines delin-
eate the threshold used with the AD algorithm. In addition, provisional regions
areallowed fortheeastlimbwherethemagneticinformationbecomessparsedue
to the time difference between observations that constitute the average magne-
togram. The threshold for the percent unipolarity is lowered for regions within
4 Henney & Harvey
es es
ol ol
al H 1000 al H 100
n n
o o
or or
C C
of of
er er
mb 100 mb
Nu Nu 10
-50 0 50 -50 0 50
Latitude (deg) CMD (deg)
1.2 1.2
W 1.0 W 1.0
Q Q
E 0.8 E 0.8
d d
ze 0.6 ze 0.6
mali 0.4 mali 0.4
or or
N 0.2 N 0.2
0.0 0.0
-50 0 50 -50 0 50
Latitude (deg) CMD (deg)
y 100 y 100
arit 95 arit
ol ol 90
nip 90 nip
U U
nt 85 nt 80
e e
erc 80 erc
P P 70
n 75 n
a a
e e
M 70 M 60
-50 0 50 -50 0 50
Latitude (deg) CMD (deg)
Figure 2. Coronal hole parameters, for the period between 1992 through
2003, verses latitude (left) and central meridian distance (CMD, right). The
total number of holes (top), normalized helium 1083 nm equivalent width
(EQW, middle), and mean percent unipolarity (bottom) are shown. The 1-σ
errorfor the values is delineated by the verticallines. See text for discussion.
the central meridian distance band between -90 and -50 degrees (depicted by
the dotted line in Figure 2).
3. Automated Detection Recipe
The coronal hole detection process begins with a two-day average He I 1083 nm
spectroheliogram and a two-day average photospheric magnetogram. The aver-
age images are created as discussed in the previous section. The helium image
averaging assists in the detection of coronal holes by taking advantage of the
low intrinsic network contrast within coronal holes (e.g. Harvey & Recely 2002).
The weighted average images are trimmed on the east and west limb regions
for unsigned central meridian distances greater than 70 degrees (see Figure 3a
and 3b).
Automated Coronal Hole Detection 5
Figure 3. Selected images throughout the automated coronal hole detec-
tion process are depicted: a) the initial average helium spectroheliogram in
heliographiccoordinates;b)theaveragemagnetogram;c)afterretainingonly
valuesfromtheimageshownina)above0.10ofthemedianofthepositiveval-
ues;d)afterapplyingtheClosefunction;e)afterapplyingtheOpenfunction;
f) after removing small and low percentage unipolarity regions.
6 Henney & Harvey
NSO/KP CORONAL HOLE MAP: HE I 1083 nm
Preliminary N Bo=_7.2
E W
03/12/03 1748 UT
S
Figure4. Anexamplewhenthe HD(bottom, left) andAD(bottom, right)
coronal hole images disagree. KPVT He I 1083 nm spectroheliogram (top,
left), and EIT 19.5 nm Fe XII emission line image (top, right) observed on
March 12, 2003 are also shown.
Starting with the average helium image, the first step is to retain only val-
ues above 1/10 the median of positive values (based on the mean values shown
in Figure 2). After removing the negative values, the candidate coronal hole
regions become clearly visible (see Figure 3c). The next two steps are to spa-
tially smooth the remaining positive values and set all positive value pixels to
a fixed value. The resultant image is then smoothed with the morphological
image analysis function Close (e.g. Michielsen & De Raedt 2001) using a square
kernel function as the shape-operator. Shown in Figure 3d, the Close function
fills the gaps and connects associated or nearby regions to define the areas of
the coronal hole candidates. The regions that are too small to be coronal holes
arethenremoved (see Harvey & Recely 2002). Thespatial imagepixelpositions
of non-zero spatial points of the resultant image (e.g. Figure 3d) are used to
fill the corresponding pixels in the image that was created after the first step
4
(e.g. Figure 3c) with large values (e.g. 10 ). After being spatially smoothed,
the large values result in the filling of small gaps and holes. The resultant
Automated Coronal Hole Detection 7
Figure5. SubregionofanautomatedcoronalholesynopticmapforCarring-
ton rotation 2005. The gray-scale reflects the polarity (positive and negative
are represented by white and black respectively) and the number of coronal
hole detections at a given latitude and longitude. The central region is the
same coronal hole exhibited in Figure 1 and 3.
image is then smoothed with the morphological image analysis function Open
(e.g. Michielsen & De Raedt 2001), where the Open function removes small fea-
tures while preserving the size and shape of larger regions (see Figure 3e). For
more discussion on morphological image analysis and standard algorithms for
segmenting images see Ja¨hne (2004).
The primary goal of the steps so far has been to consolidate and define
the boundaries of candidate coronal hole regions. The next step is to use the
average magnetic image data to determine the percentage of unipolarity of each
candidate region. Candidate regions that have a percentage of unipolarity lower
than the thresholds (delineated by the dashed lines in Figure 2) are removed.
Thefinalstep is tonumberthecoronal holecandidates andapply thesign of the
region polarity (e.g. Figure3f). Candidateregions withmean unipolarity values
that fall within the dashed and dotted lines depicted in Figure 2 are considered
provisional. These regions are distinguished in the final output image by adding
a fractional value (e.g. 0.5) to the region number value of the image pixels for
the given coronal hole area.
3.1. Coronal Hole Map Comparison
IngeneralthecoronalholecandidateregionsdetectedbytheADalgorithmagree
well with the HD images. For the 11 years compared, the automated coronal
holeareashaveapproximately3%lessareathanthehand-drawnregionsbetween
-50 through 90 degrees latitude. The difference increases for the latitude band
between -90 and -50 degrees. However, the increase is believed to be the result
of incomplete observations used in the average images (i.e. the KPVT full-disk
observations always began in the north). Comparisons between the HD and AD
coronal hole images are shown in Figures 1 and 4. The two images agree for the
July 14, 2003 example shown in Figure 1. However, comparisons with only the
HD maps have the limitation of not comparing directly with coronal measure-
8 Henney & Harvey
ments. Figure 4 highlights this difficulty. For this example, the AD map is in
better agreement with the EIT image. A detailed comparison with other coro-
nalhole detection methods (e.g. de Toma & Arge2005;Malanushenko & Jones
2005) is planned. These comparisons will undoubtedly result in further refine-
ment to the presented AD algorithm.
3.2. Coronal Hole Synoptic Maps
In addition to creating coronal hole images, these individual images are com-
bined to create synoptic maps for fixed Carrington rotation or daily synoptic
maps. Each pixel value of the synoptic map represents the number of detections
that a coronal hole is observed for that location. A subregion from an example
Carrington map is shown in Figure 5. These maps reflect the estimated coro-
nal hole boundary variations as a result of temporal evolution of the region in
addition to the quality of the detection. The new synoptic maps will be pub-
licly available via the NSO digital library along with the daily AD coronal hole
images.
4. Acknowledgments
Thecoronal holedata usedherewas compiled by K.Harvey and F.Recely using
National Solar Observatory (NSO) KPVT observations under a grant from the
NationalScienceFoundation(NSF).NSOKittPeakdatausedhereareproduced
cooperatively by NSF/AURA,NASA/GSFC, andNOAA/SEL.TheEITimages
are courtesy of the SOHO/EIT consortium. This research was supported in
part by the Office of Naval Research Grant N00014-91-J-1040. The NSO is
operated by the Association of Universities for Research in Astronomy, Inc.
under cooperative agreement with the NSF.
References
Andretta, V., & Jones, H. P. 1997, ApJ, 489, 375
Arge, C. N., Luhmann, J. G., Odstrcil, D., Schrijver, C. J., Li, Y. 2004, Journal of
Atmospheric and Solar-TerrestrialPhysics, 66, 1295
Bohlin, J. D. 1977, in Coronal Holes and High Speed Wind Streams, ed. J. B. Zirker,
(Colorado:CAU Press), 27
Harvey, J. W., & Sheeley, N. R., Jr. 1977,Solar Physics, 54, 343
Harvey, J. W., & Livingston, W. C. 1994, in IAU Symp. 154, Infrared Solar Physics
(Dordrecht:Kluwer), 59
Harvey, K. L., & Recely, F. 2002, Solar Physics, 211, 31
Henney, C. J., & Harvey, J. W. 2001, at the American Geophysical Union, Spring
Meeting, SP51B-03
J¨ahne, B. 2004, Practical Handbook on Image Processing for Scientific and Technical
Applications, 2nd ed., (Boca Raton:CRC Press)
Malanushenko, O. V., & Jones, H. P. 2005, Solar Physics, in press
Michielsen, K., & De Raedt, H. 2001,Physics Reports, 347, 461
de Toma, G., & Arge, C. N., these Proceedings
Zirker, J. B. 1977, in Coronal Holes and High Speed Wind Streams, ed. J. B. Zirker,
(Colorado:CAU Press), 1